Method, system, equipment and medium for geometric description of small defects on high-speed train tread

Abstract

The present application belongs to the field of rail transit safety monitoring technology, and discloses a method, system, equipment and medium for geometric description of tiny defects on high-speed train treads. The method comprises: collecting morphological feature data of tiny defects on high-speed train wheel treads, grouping and extracting key features, and constructing an original defect geometry database; establishing a transient finite element model of wheel-rail rolling contact, simulating the mechanical response of defects under different working conditions, and outputting stress cloud maps and stress intensity factors; defining a working condition parameter matrix, calculating the defect expansion rate, extracting defect expansion characteristics, and using a deep autoencoder and a conditional generative adversarial network to generate a defect instability expansion morphological feature set; using a convolutional neural network to construct a mapping relationship model and an inversion model between rail strain signals and geometric parameter vectors; obtaining real-time rail strain signals, inputting the inversion model to predict the geometric parameter vectors, and accurately generating a geometric description of tiny defects.

Description

Technical Field

[0001] The present invention relates to the technical field of rail transit safety monitoring and intelligent diagnosis, and in particular to a method, system, equipment and medium for geometrically describing tiny defects on a high-speed train tread. Background Art

[0002] High-speed trains are a vital pillar of the modern transportation system. Their operational safety and reliability are crucial for protecting passenger safety and improving transportation efficiency. As the core component of a train that directly contacts the rails, the condition of wheelsets directly determines the train's operational stability and safety. Currently, high-speed train wheelsets are primarily maintained using a planned maintenance system, with maintenance intervals determined based on mileage or age, and assigned to five maintenance levels. This system aims to ensure train safety through regular maintenance, but it has shown limitations in practice. Specifically, monitoring the condition of the wheel tread, a key area of wheel-rail contact, is particularly important and is typically included in daily inspections. After high-speed trains enter depot, personnel visually inspect the wheel treads to assess damage. However, tracking tests have revealed that high-speed train wheel treads are often damaged by minor defects such as rubbing, particularly on leading and trailing wheels. Due to the complex forces acting on them, these early minor defects, if not promptly detected and addressed, can gradually develop into instability (such as tread delamination), threatening the operational safety of the wheel-rail system.

[0003] The existing technology has the following significant defects and deficiencies in the detection and treatment of small defects in high-speed train wheel treads:

[0004] 1. Limitations of visual inspection and risk of missed inspections: Currently, routine inspections of wheel tread condition rely primarily on visual inspection by personnel. This method is limited by visual blind spots, making minor defects easily overlooked, especially in low-light conditions or complex environments. Furthermore, visual inspections cannot scientifically assess the morphological characteristics of defects or their expansion trends, leading to the failure to promptly detect potential safety hazards and increasing driving risks.

[0005] 2. Random and inefficient decisions regarding temporary lathing: When minor defects are discovered during visual inspection, the decision to perform temporary lathing, as well as the depth and frequency of lathing, often depends on the operator's personal experience, lacking a scientific basis for decision-making. Because existing technology cannot accurately determine the depth and internal morphology of the defect, temporary lathing operations are highly arbitrary, resulting in low efficiency and difficulty ensuring lathing quality, further impacting the service life of the wheel. Summary of the Invention

[0006] The purpose of the present invention is to provide a method, system, equipment and medium for geometric description of small defects on the tread of a high-speed train. Through parametric modeling and transient simulation technology, accurate geometric description and mechanical response analysis of small defects on the wheel tread can be achieved, so as to improve the scientificity and accuracy of defect identification.

[0007] In a first aspect, the present invention provides a method for geometrically describing small defects on a high-speed train tread, comprising the following steps:

[0008] S1: Collecting morphological feature data of tiny defects on the wheel tread of a high-speed train to construct an original defect feature dataset, grouping the defects in the original defect feature dataset using aspect ratio; extracting key morphological features from each group of defects to construct a characteristic parameter matrix of the tiny defects on the wheel tread; constructing a geometric function of the tiny defects on the wheel tread using piecewise polynomial fitting and cubic spline interpolation; and constructing a standardized original defect geometry database matrix based on the characteristic parameter matrix and the geometric parameter vector in the geometric function;

[0009] S2: Establish a transient finite element model of high-speed train wheel-rail rolling contact, import the geometric function into the transient finite element model, and define the wheel-rail contact relationship; use the penalty function method to define the wheel-rail contact constraint, and set the tangential friction in combination with the Coulomb friction model, and use an adaptive mesh in the defect area; perform mechanical response simulation of the transient finite element model under different operating conditions, output the Mises stress in the defect area, and calculate the stress intensity factor using the J-integral method;

[0010] S3: defining an operating condition parameter matrix, the operating condition parameter matrix including the geometric parameter vector; approximating the defect growth rate based on the stress intensity factor and the Paris law; extracting the maximum Mises stress, maximum stress intensity factor, and average defect growth rate for each operating condition to construct a defect growth characteristic; based on the defect growth characteristic, using a deep autoencoder and a conditional generative adversarial network to generate a feature set of defect instability growth morphology;

[0011] S4: collecting rail strain signals through multiple FBG sensors; extracting time domain, frequency domain, and time-frequency domain features of the rail strain signals to generate a first standardized feature matrix; using a convolutional neural network to construct a mapping relationship model between the first standardized feature matrix and the geometric parameter vector, and developing a convolutional neural network inversion model;

[0012] S5: Acquire the real-time rail strain signal and convert it into a second standardized feature matrix; input the second standardized feature matrix into the trained convolutional neural network inversion model to obtain a predicted geometric parameter vector and perform weighted correction; based on the weighted corrected predicted geometric parameter vector, obtain a three-dimensional predicted geometric description according to the geometric function.

[0013] As an optional implementation of the first aspect of the present application, in step S1, the morphological feature data includes the depth, area, width, length, curvature and edge gradient of the defect.

[0014] As an optional implementation of the first aspect of the present application, in step S1, the geometric function is obtained by performing piecewise polynomial fitting on the defect cross-sectional profile and using cubic spline interpolation in the edge transition region.

[0015] As an optional implementation of the first aspect of the present application, in step S2, the transient finite element model is established using ABAQUS software, and the dynamic update of the defect geometry is achieved through a Fortran subroutine.

[0016] As an optional implementation of the first aspect of the present application, in step S3, the deep autoencoder is used to extract potential features from the defect expansion characteristics, and the conditional generative adversarial network uses the potential features and operating parameters to generate extended feature samples, and then identifies the unstable expansion pattern through a clustering algorithm to form the defect unstable expansion morphological feature set.

[0017] As an optional implementation of the first aspect of the present application, in step S4, the time domain features include peak strain, root mean square strain and kurtosis; the frequency domain features include peak frequency and spectral energy; and the time-frequency domain features are used to extract wavelet energy through wavelet transform.

[0018] As an optional implementation of the first aspect of the present application, in step S5, the weighted correction is to weightedly fuse the geometric parameter vector predicted by the convolutional neural network inversion model with the cluster center of the defect instability expansion morphological feature set.

[0019] In a second aspect, an embodiment of the present application provides a high-speed train tread micro-defect geometric description system, comprising:

[0020] The data acquisition and preprocessing module is configured to collect morphological feature data of micro defects on the wheel tread of a high-speed train, construct an original defect feature data set, group the defects in the original defect feature data set using aspect ratio; extract key morphological features from each group of defects to construct a characteristic parameter matrix of the micro defects on the wheel tread; construct a geometric function of the micro defects on the wheel tread using piecewise polynomial fitting and cubic spline interpolation; and construct a standardized original defect geometry database matrix based on the characteristic parameter matrix and the geometric parameter vector in the geometric function;

[0021] a finite element simulation module configured to establish a transient finite element model of high-speed train wheel-rail rolling contact, import the geometric function into the transient finite element model, and define the wheel-rail contact relationship; define the wheel-rail contact constraint using a penalty function method, set tangential friction in combination with a Coulomb friction model, and use an adaptive mesh in the defect area; perform mechanical response simulation of the transient finite element model under different operating conditions, output the Mises stress in the defect area, and calculate the stress intensity factor using a J-integral method;

[0022] a defect growth characteristic analysis and feature set generation module configured to define an operating condition parameter matrix, the operating condition parameter matrix including the geometric parameter vector; approximate the defect growth rate based on the stress intensity factor in combination with the Paris law; extract the maximum Mises stress, maximum stress intensity factor, and average defect growth rate for each operating condition to construct a defect growth characteristic; and generate a defect instability growth morphology feature set based on the defect growth characteristic using a deep autoencoder and a conditional generative adversarial network;

[0023] a signal acquisition and inversion modeling module configured to acquire rail strain signals through a plurality of FBG sensors; extract features of the rail strain signals in the time domain, frequency domain, and time-frequency domain to generate a first standardized feature matrix; construct a mapping relationship model between the first standardized feature matrix and the geometric parameter vector using a convolutional neural network, and develop a convolutional neural network inversion model;

[0024] The real-time geometric description module is configured to obtain a real-time rail strain signal and convert it into a second standardized feature matrix; input the second standardized feature matrix into a trained convolutional neural network inversion model to obtain a predicted geometric parameter vector and perform weighted correction; based on the weighted corrected predicted geometric parameter vector, obtain a three-dimensional predicted geometric description according to the geometric function.

[0025] In a third aspect, an embodiment of the present application provides an electronic device comprising a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein when the program or instruction is executed by the processor, the steps of the method described in the first aspect are implemented.

[0026] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented.

[0027] Compared with the existing technology, the present invention first collects the morphological characteristic data of small defects on the wheel tread, groups the defects using the aspect ratio, and constructs a geometric function by combining piecewise polynomial fitting and cubic spline interpolation, thereby achieving accurate parametric modeling and three-dimensional geometric description of small defects on the wheel tread. This directly solves the problem that visual inspection cannot scientifically evaluate the morphological characteristics of defects. Secondly, the present invention establishes a transient finite element model of high-speed train wheel-rail rolling contact, imports the constructed geometric function for mechanical response simulation, and uses the penalty function method to define contact constraints and the Coulomb friction model to set tangential friction. At the same time, an adaptive grid is used in the defect area to output Mises stress and calculate the stress intensity factor. It can quantitatively analyze the mechanical behavior and damage degree of the defect area under different operating conditions, thereby making up for the deficiency of the existing technology that cannot deeply grasp the defect depth and internal morphological characteristics. Furthermore, the present invention approximates the defect expansion rate based on the stress intensity factor combined with the Paris law, and utilizes a deep autoencoder and a conditional generative adversarial network to generate a set of defect instability expansion morphological features, achieving a scientific prediction of the future expansion trend of the defect and an assessment of the instability risk. This provides a quantitative scientific basis for optimizing temporary turning and repair decisions, and avoids arbitrary decision-making. Finally, the present invention uses multiple FBG sensors to collect rail strain signals in real time, and uses a convolutional neural network to construct a mapping relationship model between the rail strain signal and the defect geometric parameter vector and its inversion model, thereby obtaining and predicting the three-dimensional geometric description of the defect in real time. This innovative method completely overcomes the shortcomings of visual inspection, which is limited by visual blind spots, prone to missed inspections, and low efficiency, and realizes non-contact, highly efficient real-time defect detection and assessment. Overall, the multi-level technical means adopted in the present invention, such as parametric modeling, transient simulation, model prediction and real-time monitoring, have solved the core problems of unscientific identification of minor defects on the treads of high-speed train wheels, high risk of missed detection, and lack of basis for maintenance decisions. It has significantly improved the scientificity, accuracy and real-time nature of defect identification, reduced the driving safety risks caused by the expansion of early minor defects, and effectively improved the service safety and maintenance efficiency of train wheelsets. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flow chart of a method for geometrically describing small defects on a high-speed train tread according to an embodiment of the present invention;

[0029] Figure 2 This is the structure diagram of the high-resolution image moiré removal model based on pyramid feature extraction and attention feature fusion;

[0030] Figure 3 is a structural diagram of an encoder / decoder according to an embodiment of the present invention;

[0031] Figure 4is a structural diagram of a dilated residual dense block (DRDB) according to an embodiment of the present invention;

[0032] Figure 5 is a structural diagram of an enhanced attention gate (EAG) according to an embodiment of the present invention;

[0033] Figure 6 It is a structural schematic diagram of a high-speed train tread micro-defect geometric description system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0035] The terms "first", "second", etc. in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the application can be implemented in a sequence other than those illustrated or described here. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally represents that the objects associated before and after are in a kind of "or" relationship. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically limited.

[0036] Example 1

[0037] See also Figure 1 , which is a flow chart of a method for geometric description of tiny defects on a high-speed train tread provided by an embodiment of the present invention. The method includes five stages: original defect data acquisition and modeling stage, wheel-rail contact simulation and mechanical response analysis stage, defect expansion characteristic generation and morphological feature set construction stage, strain signal processing and recognition model construction stage, and real-time recognition and geometric description application stage. The specific implementation process is as follows.

[0038] Phase 1: Original defect data collection and modeling. This phase can be divided into the following steps:

[0039] I. Collecting morphological feature data of tiny defects on the wheel tread of a high-speed train, constructing an original defect feature dataset, and grouping the defects in the original defect feature dataset using aspect ratio.

[0040] During the service life of high-speed train wheels, various testing equipment (wheel profile detector, wheel out-of-roundness detector, metal surface micro-defect detector, laser ultrasonic damage detector, steel ruler, etc.) are regularly used to collect morphological characteristic data of micro-defects on the wheel tread. Let t represent the service time (unit: day) and m represent the service mileage (unit: kilometer). For each defect , the morphological features at time t are characterized by the feature vector express:

[0041] ;

[0042] The definitions of each component are as follows:

[0043] : Depth of defect i at time t (unit: mm);

[0044] : the area of defect i at time t (unit: mm2);

[0045] : the width of defect i at time t (unit: mm);

[0046] : length of defect i at time t (unit: mm);

[0047] : The curvature of defect i at time t (unit: mm -1 ), reflecting the curvature of the defect surface;

[0048] : The edge gradient of defect i at time t (dimensionless), indicating the steepness of the edge.

[0049] Then construct the original defect feature dataset:

[0050] ,

[0051] in: is the total number of defects, is the total number of measurement time points.

[0052] To group defects, define the aspect ratio Based on this, the defects are divided into Groups ( is the total number of groups), each group corresponds to an aspect ratio range , the set of defects of group k is expressed as:

[0053] .

[0054] II. Extract key morphological features from each group of defects to construct a characteristic parameter matrix of the wheel tread micro-defects.

[0055] For each set of defects , extract its key morphological feature parameters and construct the feature parameter matrix , let the kth group contain eigenvectors, the matrix is defined as:

[0056] ,

[0057] in, is a A three-dimensional matrix with each row corresponding to an eigenvector , ;

[0058] is the total number of feature vectors of the kth group of defects; is the time point corresponding to the nth eigenvector.

[0059] : is the nth defect in the kth group at time The eigenvalues of .

[0060] Analyze the variation of characteristic parameters with service mileage m, and for each parameter , described by a polynomial regression model:

[0061] ,

[0062] in, , ,…, is the regression coefficient; n is the polynomial order; is a random error term. The vector form is defined as: coefficient vector: ; Independent variable vector: The prediction model is:

[0063] ,

[0064] Solve using the least squares method The observation dataset is , construct the matrix:

[0065] ;

[0066] Then the regression coefficient solution can be obtained as: .

[0067] III. Using piecewise polynomial fitting and cubic spline interpolation method, a geometric function of the wheel tread micro-defects is constructed.

[0068] In order to describe the geometric shape of the defect, piecewise polynomial fitting and cubic spline interpolation are used. Assume that the defect cross-sectional profile consists of S segments, each segment is represented by a polynomial:

[0069]

[0070] Among them: s=1,2,…,S represents the total number of segments, represents the position of the sth node; P is the polynomial order; is the p-th order coefficient of the s-th segment, Indicates the horizontal coordinate of the contour of Power is used to construct polynomial functions to describe profile characteristics;

[0071] To ensure smoothness, adjacent segments must meet the continuity condition:

[0072] ;

[0073] In the edge transition area, cubic spline interpolation is used. Let the interpolation node be , cubic spline function satisfy:

[0074] ,

[0075] in is the horizontal coordinate of the interpolation node, is the corresponding depth value, Indicates the number of interpolation nodes.

[0076] The defect geometry model is a parameterized function:

[0077] ,

[0078] in: , represents the total parameter vector composed of the coefficient vectors of all segments, with a dimension of ; is the coefficient vector of the s-th polynomial.

[0079] Expressed as the contour function of the sth segment.

[0080] IV. Constructing a standardized original defect geometry database matrix based on the characteristic parameter matrix and the geometric parameter vector in the geometric function.

[0081] Based on the above geometric function , for each defect , is the total number of defects), generating geometric instances

[0082]

[0083] Indicates the The horizontal axis interval of the segment, ,satisfy ;

[0084] Indicates defect i The first segment of the polynomial Order coefficient;

[0085] A vector of geometric parameters representing defect i.

[0086] Combined with its characteristic parameter matrix (See Part II of Phase I for details) to construct the original defect geometry database matrix:

[0087] ,

[0088] in, The characteristic parameter matrix representing defect i has a dimension of M × 6 and contains all the characteristic vectors of the k-th defect at the M-th time point;

[0089] The transpose of the geometric parameter vector representing defect i;

[0090] Represents the original defect geometry database matrix, with dimensions of , integrating the geometric parameters and characteristic parameters of all defects.

[0091] For the convenience of analysis, the characteristic parameter matrix After normalization, the original defect geometry database matrix is:

[0092] .

[0093] Phase 2: Wheel-rail contact simulation and mechanical response analysis. This phase can be divided into the following steps:

[0094] I. Establish a transient finite element model of high-speed train wheel-rail rolling contact, import the geometric function into the transient finite element model, and define the wheel-rail contact relationship.

[0095] A transient finite element model of high-speed train wheel-rail rolling contact was established using ABAQUS to simulate wheel-rail contact behavior and defect propagation. The wheel and rail geometric models were created in ABAQUS, with the rail cross-section based on the actual standard (UIC60). After defining the wheel-rail contact relationship, an initial finite element model was generated and exported as a .inp file. Based on field tracking test data, the initial wheel tread defect was defined in the .inp file using the *IMPERFECTION keyword. The defect geometry was represented by a parameterized function, and the defect geometry function constructed in Phase 1 was imported. and the normalized original defect geometry database matrix ,as follows:

[0096] from Extract each defect Characteristic parameters , defining the parameter vector of typical defects:

[0097] ,

[0098] in, represent the depth, area, width, length, curvature and edge gradient of the i-th defect respectively.

[0099] The defect surface height function is:

[0100] ,

[0101] in, are the polynomial coefficients of the sth segment, which are called from the database in stage 1.

[0102] In ABAQUS, the dynamic update of defect geometry is realized by using the Fortran subroutine (USDFLD). and , adjust the tread node coordinates:

[0103] ,

[0104] in, is the original node coordinate, and n is the normal vector.

[0105] II. The penalty function method is used to define the wheel-rail contact constraint, and the tangential friction is set in combination with the Coulomb friction model. An adaptive grid is used in the defect area.

[0106] The boundary conditions of wheel-rail rolling contact need to consider the influence of the defect geometry in stage 1. The contact constraints are defined using the penalty function method: ,

[0107] in, is the normal penalty function stiffness, is the contact gap, , are the surface coordinates of the wheel and rail respectively, and the gap is directly affected by Influence;

[0108] The tangential friction is set based on the Coulomb friction model: , where μ=0.3 is the friction coefficient, is the tangential penalty function stiffness, is the tangential relative velocity. Adaptive grid is used in the defect area, and the unit size is determined by the defect width of stage 1. control:

[0109] ,

[0110] in Indicates the minimum grid size of the defect area, which is determined by the first stage Sure, is the mesh scaling factor to ensure accurate capture of the local stress field of the defect.

[0111] III. Perform mechanical response simulation of the transient finite element model under different operating conditions and output the Mises stress of the defect area.

[0112] The load is applied to the wheel center, combined with the defect position of stage 1 (from ), simulating the dynamic response of the train passing through the defect. Transient finite element simulation is performed in ABAQUS / Explicit, using the defect parameter vector provided in stage 1. , geometric description function and the standardized matrix , simulates the mechanical response of small defects on the wheel tread under different operating conditions. The output result is the Mises stress cloud diagram of the wheel (such as Figure 2 ) and stress intensity factors (as shown in Figure 3 As shown in Figure 2, the defect expansion trend is evaluated. The details are as follows:

[0113] From the standardized defect database Extract , corresponding to the defect instance of a specific working condition ,Will Input to the ABAQUS Fortran subroutine, by calculating Update the coordinates of the mesh nodes in the defect area.

[0114] Through the Fortran subroutine Surface height distribution converted to defect area , used to update the wheel tread geometry.

[0115] from Extract each defect Characteristic parameters , used to define the operating conditions. According to the service mileage, select the corresponding standardized feature vector .

[0116] Will Denormalized to physical units, with Input the simulation model together to determine the geometric size and boundary conditions of the defect. Based on N defect instances and M time points, multiple simulation conditions are generated to cover different defect forms and service conditions.

[0117] Then, the explicit dynamics method is used to solve the transient process, and the control equation is:

[0118] ,

[0119] in, is the wheel material density; displacement field; is the stress tensor; External force.

[0120] Time discretization uses the central difference method, and the time step Satisfy the stability conditions:

[0121] ,

[0122] in, represents the minimum grid size of the defect area; c represents the longitudinal wave velocity of the material; =210GPa, which is the Young's modulus of the wheel material.

[0123] Calculate the Von Mises stress distribution on the wheel tread and defect area as the output of the stress contour map:

[0124] ,

[0125] Indicates the location (three-dimensional space coordinates, unit: millimeter, mm) and time The Von Mises stress at (unit: MPa) represents the equivalent stress of the material and is used to assess whether the yield limit has been reached.

[0126] in: is the deviatoric stress tensor; it is defined as:

[0127] ,

[0128] Where, is the stress tensor, which contains the components , through the constitutive relation Calculation; C is the elastic stiffness tensor, is the total strain, is the plastic strain, represents the unit tensor ( identity matrix);

[0129] is the volumetric stress tensor (unit: MPa), a The symmetry tensor, representing the volumetric component of the stress tensor (i.e., the isotropic compressive or tensile part), is defined as:

[0130] ,

[0131] In ABAQUS, the Mises stress cloud diagram is generated by the post-processing module and stored as a spatial distribution matrix:

[0132] ,

[0133] in, is the coordinate of the wheel tread mesh node; is the simulation time step; N is the total number of nodes; M is the total number of measurement time points.

[0134] IV. Calculate the stress intensity factor using the J-integral method.

[0135] For defects defined in stage 1, the stress intensity factor K is calculated to evaluate the driving force for expansion. The J-integral method is used for calculation:

[0136] ,

[0137] in: , is the strain energy density; , is the surface traction; is the integral path around the defect tip; is the path infinitesimal element.

[0138] For a plane stress state, the stress intensity factor The relationship with J integral is

[0139] ,

[0140] In ABAQUS, the *CONTOUR INTEGRAL function is used to extract , for each defect instance in phase 1 , generate a time series vector:

[0141] .

[0142] In summary, the Mises stress cloud diagram is Output in the form of vector for visual analysis of stress concentration in defect area. Formal output, characterizing the driving force of defect expansion.

[0143] Phase 3: Defect expansion feature generation and morphological feature set construction. This phase can be divided into the following steps:

[0144] I. Define an operating condition parameter matrix, wherein the operating condition parameter matrix includes the geometric parameter vector.

[0145] In order to fully simulate the defect growth characteristics under different working conditions, a working condition parameter matrix is defined, covering key variables such as train speed, load and track surface conditions.

[0146] ,

[0147] in, is the defect geometry parameter vector corresponding to the kth working condition, from the standardized defect geometry database of stage 1 Extract and describe the contour shape of the defect, is the total number of defect contour segments.

[0148] is the train speed (unit: m / s, );

[0149] is the axle load (unit: N, );

[0150] is the height of rail surface irregularity;

[0151] : total number of working conditions;

[0152] II. Based on the stress intensity factor, the defect growth rate is approximately calculated in combination with the Paris law.

[0153] Relying on ABAQUS simulation software, based on the transient finite element framework of the second stage, the defect expansion characteristics under different working conditions are simulated in batches: for each working condition k, Enter the dynamic load timing of stage 2:

[0154] ,

[0155] Where A is the amplitude coefficient, is the track irregularity wavelength, and t is the simulation time. The defect geometry is given by Definition: Use Python scripts in ABAQUS to automatically execute N simulations, outputting Mises stress contours and stress intensity factors for each calculation.

[0156] The defect growth rate is approximately calculated based on Paris's law:

[0157] ,

[0158] in, is the mode I stress intensity factor of the kth load case at time step t, 、 are material constants. The simulation results are stored as an extended characteristic matrix:

[0159] , .

[0160] III. For each operating condition, the maximum Mises stress, maximum stress intensity factor, and average defect growth rate are extracted to construct the defect growth characteristics.

[0161] For each load case k, the maximum Mises stress is extracted (e.g. Figure 4 shown), maximum stress intensity factor and average expansion rate:

[0162] ,

[0163] in, ; ; .

[0164] IV. Based on the defect expansion characteristics, a deep autoencoder and a conditional generative adversarial network are used to generate a feature set of defect instability expansion morphology.

[0165] The characteristics of all working conditions are as follows: , further, based on , a method combining deep autoencoder (DAE) and conditional generative adversarial network (CGAN) is used to generate a morphological feature set of unstable expansion of small defects. The definition of DAE includes encoder and decoder:

[0166] Encoder: ;

[0167] Decoder: ;

[0168] in, is the input feature vector; For potential representation; is the weight matrix; is the bias vector; is the activation function that minimizes the reconstruction error:

[0169] ,

[0170] in, is the regularization parameter; the potential feature matrix is obtained by Adam optimizer ;

[0171] Then, CGAN uses Z and working condition parameters Generate extended feature samples,

[0172] The generator is defined as: ;

[0173] The discriminator is: ;

[0174] Where, 、 are the neural network parameters of the generator and discriminator;

[0175] The loss function uses conditional GAN adversarial loss as:

[0176] ,

[0177] Represents a random variable According to its probability distribution Take expectations; is the probability distribution of the real feature, representing Statistical properties of the eigenvectors in .

[0178] represents a random variable According to its probability distribution Combined expectations, is the probability distribution of the underlying noise, typically a standard normal distribution, used to provide random input to the generator.

[0179] The goal is to minimize ,maximize ;

[0180] By alternately optimizing the generator and discriminator (learning rate , iteration ), generate feature set:

[0181] ,

[0182] M is the number of generated samples, and the real features and generate features merge , using DBSCAN algorithm for density clustering (neighborhood radius , the minimum number of samples MinPts=5), identify K unstable expansion modes, and the cluster center is

[0183] ,

[0184] in, is the kth cluster center, is the total number of clusters.

[0185] The final unstable expansion morphological feature set is defined as:

[0186] .

[0187] in, It is a comprehensive feature set, including real features, generated features and cluster centers, which is used to characterize the morphological pattern of defect instability.

[0188] Phase 4: Strain signal processing and identification model construction. This phase can be divided into the following steps:

[0189] Ⅰ. Collect rail strain signals through multiple FBG sensors.

[0190] In order to achieve high-precision and real-time rail strain monitoring, a signal acquisition system based on FBG is designed and built. The bottom of the rail is deployed along the wheel-rail contact path. FBG sensors, whose positions are represented by the matrix:

[0191] ,

[0192] in, represents the jth sensor ( ) and the defect center in stage 2 Alignment, are the horizontal and vertical coordinates along the rail respectively, and the sensor spacing is , represents the length of the wheel defect at time t, from the standardized defect geometry database of stage 1 The defect parameter vector in extract.

[0193] The relationship between the wavelength shift of the FBG sensor and the strain and temperature is:

[0194] ,

[0195] Where: is the wavelength shift of the jth sensor, is the initial wavelength of the jth sensor, =0.22 is the photoelastic coefficient, is the local strain, is the coefficient of thermal expansion, is the temperature change; the local strain of the jth sensor is simplified to:

[0196] ,

[0197] Data collection part, set the collection frequency , record the strain signal in the time interval [0,T] and generate a time series matrix:

[0198] ;

[0199] in, is the kth sampling time point, , the complete strain signal matrix is:

[0200] ,

[0201] Where, For time The strain vector transpose.

[0202] II. Extracting the rail strain signal in time domain, frequency domain and time-frequency domain to generate a first standardized feature matrix.

[0203] from The sensitivity characteristics caused by small defects are extracted and quantified, and the time domain, frequency domain and time-frequency domain analysis methods are used to fully characterize the defect response. In order to capture the strain anomaly caused by defect expansion, the time domain characteristics of the jth sensor are calculated, including peak strain, root mean square strain and kurtosis, where the peak strain, root mean square strain and kurtosis are:

[0204] , , ,

[0205] in, , represents the average strain of the jth sensor;

[0206] is the jth sensor at time The strain value;

[0207] In frequency domain analysis, Perform fast Fourier transform (FFT) to obtain the spectrum , extract the peak frequency and spectrum energy, where the peak frequency of the jth sensor is: , responsible for reflecting the main frequency component of the signal;

[0208] The spectrum energy of the jth sensor is: , responsible for reflecting the frequency distribution intensity of the signal.

[0209] In order to further explore the dynamic characteristics, wavelet transform (WT) is used with Morlet wavelet as the basis function to calculate the wavelet coefficient matrix :

[0210] ,

[0211] Where, is the Morlet wavelet function, a is the scale, b is the time offset, For my mother Xiaobo, is the complex conjugate of the Morlet wavelet. And extract the wavelet energy:

[0212] ,

[0213] Combining these features, the feature vector of the jth sensor is defined ( ):

[0214] ;

[0215] Then the characteristic matrix of all sensors is ( For the sensor feature vectors):

[0216] ,

[0217] To eliminate the dimension difference, Perform Z-score normalization:

[0218] ,

[0219] , ,

[0220] Generate the standardized feature matrix (i.e. the first standardized feature matrix): , contains standardized characteristics of all sensors.

[0221] III. Use a convolutional neural network to construct a mapping relationship model between the first standardized feature matrix and the geometric parameter vector, and develop a convolutional neural network inversion model.

[0222] In order to establish the mapping relationship between the characteristics of small defects and the characteristics of rail strain signals, the unstable expansion morphology feature set of stage three is combined and the simulation results of phase 2 , using convolutional neural network (CNN) to build a model (such as Figure 5 shown).

[0223] The training set is constructed as follows:

[0224] ,

[0225] in: represents the standardized characteristic matrix of the i-th working condition, which contains the strain signal characteristics of the rail under different loading conditions; is the defect geometry parameter vector of the i-th working condition, describing the specific characteristics of the tiny defect, and N represents the total number of working conditions, that is, the number of training samples.

[0226] To adapt to the CNN input format, Reshape into tensor form. The CNN model structure includes:

[0227] Input layer: input tensor , the dimension is .

[0228] Convolution layer: Use 32 convolution kernels (size , step size is 1), ReLU activation, output The dimension is .

[0229] Pooling layer: Maximum pooling (window is , step size is 2), output The dimension is .

[0230] Convolution layer: Use 64 convolution kernels (size , step size is 1), ReLU activation, output The dimension is .

[0231] Pooling layer: Maximum pooling (window is , step size is 2), output The dimension is .

[0232] Fully connected layer: flattened into vectors (Dimensions are , mapped to the output layer (Dimensions are ,correspond . is the total number of defect contour segments.

[0233] The loss function for CNN model training is:

[0234] ,

[0235] in, is the L2 regularization coefficient. Using Adam optimizer (learning rate ), batch size B=32, training E=5000 rounds, monitoring the validation set error. It is The L2 norm squared of the layer weight matrix;

[0236] Used to measure the difference between the predicted defect parameters and the actual defect parameters.

[0237] Furthermore, a convolutional neural network inversion algorithm is developed to Perform feature extraction and standardization to obtain a standardized matrix .

[0238] In the inversion part of the deep learning model, input To the trained CNN, get the initial prediction: .

[0239] Next, optimize and correct, combined with the third stage , calculation and clustering center The Euclidean distance ( is the total number of clusters):

[0240] ;

[0241] Select the cluster center with the smallest distance: , weighted correction:

[0242] ,

[0243] is the weighting coefficient, is the final predicted defect parameter vector.

[0244] Output confidence interval ,in, Obtained based on CNN variance estimation.

[0245] Integrate FBG acquisition system, feature extraction module and inversion algorithm into a unified platform, and output in real time The data is stored as:

[0246] .

[0247] Phase 5: Real-time recognition and geometric description application. This phase can be divided into the following steps:

[0248] Ⅰ. Obtain the real-time rail strain signal and convert it into the second standardized feature matrix.

[0249] II. Input the second standardized feature matrix into the trained convolutional neural network inversion model to obtain the predicted geometric parameter vector and perform weighted correction.

[0250] The real-time rail strain signal is obtained from the FBG system in stage 4 as the input of the inversion algorithm. The real-time acquisition time window is , the acquisition frequency is , the number of sensors is The real-time strain signal is expressed as:

[0251] ;

[0252] in, is the total number of sampling points, and the complete signal matrix is:

[0253] ;

[0254] To ensure that the signal is suitable for inversion, calculate the signal energy:

[0255] .

[0256] Based on the convolutional neural network (CNN) inversion model trained in stage 4 , predict defect parameters from real-time signals. Generate the second standardized feature matrix (including peak strain, RMS strain, kurtosis, peak frequency, spectral energy, and wavelet energy), which are directly input into the model to obtain the initial prediction:

[0257] ,

[0258] in, , respectively represent the predicted defect depth, width, length, edge curvature and edge gradient. Combined with the weighted fusion method of stage 4, the final prediction parameters are obtained. .

[0259] III. Based on the weighted corrected prediction geometric parameter vector, a three-dimensional prediction geometric description is obtained according to the geometric function.

[0260] use , based on the geometric description framework of stage one, a prediction geometric description of the defect is constructed. Stage one defines the defect surface function It is a combination of piecewise polynomial and cubic spline interpolation. The prediction geometric description function is:

[0261] .

[0262] The specific implementation is as follows. Piecewise polynomial part: Assume that the defect cross section is divided into S segments, and the polynomial form of each segment is:

[0263] ,

[0264] is the p-th order coefficient of the s-th segment polynomial ( , p is the polynomial order);

[0265] In the marginal area ( )( For defect center, is the width of the transition zone, is the predicted defect width), using cubic spline smoothing:

[0266] ;

[0267] coefficient Solved with (curvature) and (gradient) constraints and boundary continuity:

[0268] ;

[0269] The three-dimensional prediction geometry description is expanded to:

[0270] ,

[0271] in, are the coordinates of the defect in the horizontal, vertical and vertical directions along the rail, is the curvature and gradient adjustment function, which is determined by the spline coefficients and boundary conditions.

[0272] IV. Converting the three-dimensional predicted geometric description into a predicted geometric description in the form of a mesh required for finite element simulation of the defect area.

[0273] Will Convert to the mesh format required for finite element simulation. Define the mesh points in the defect area:

[0274] ,

[0275] in: is the coordinate of the grid point along the x-axis; (unit: mm, )

[0276] is the coordinate of the grid point along the y-axis (unit: mm, )

[0277] Indicates the x-axis grid resolution; Indicates the y-axis grid resolution; 、 is the number of grid points corresponding to the coordinate axis.

[0278] Calculate the height of each grid point as ( is the coordinate of the kth grid point). Generate the predicted geometric description:

[0279] ,

[0280] is the total number of grid points,

[0281] Output prediction geometry description For subsequent use, stored as structured data:

[0282] ,

[0283] is the real-time strain signal matrix; is the standardized feature matrix; is the final predicted defect parameter vector.

[0284] Example 2

[0285] See also Figure 6 , shown is a schematic structural diagram of a high-speed train tread micro-defect geometry description system proposed in the second embodiment of the present application. The system includes the following key modules:

[0286] The data acquisition and preprocessing module 100 is configured to collect morphological feature data of micro-defects on the wheel tread of a high-speed train, construct an original defect feature data set, group the defects in the original defect feature data set using aspect ratio; extract key morphological features from each group of defects to construct a characteristic parameter matrix of the micro-defects on the wheel tread; construct a geometric function of the micro-defects on the wheel tread using piecewise polynomial fitting and cubic spline interpolation; and construct a standardized original defect geometry database matrix based on the characteristic parameter matrix and the geometric parameter vector in the geometric function;

[0287] The finite element simulation module 200 is configured to establish a transient finite element model of high-speed train wheel-rail rolling contact, import the geometric function into the transient finite element model, and define the wheel-rail contact relationship; define the wheel-rail contact constraint using a penalty function method, set tangential friction in combination with a Coulomb friction model, and use an adaptive mesh in the defect area; perform mechanical response simulation of the transient finite element model under different operating conditions, output the Mises stress in the defect area, and calculate the stress intensity factor using a J-integral method;

[0288] The defect growth characteristic analysis and feature set generation module 300 is configured to define an operating condition parameter matrix, the operating condition parameter matrix including the geometric parameter vector; approximate the defect growth rate based on the stress intensity factor in combination with the Paris law; extract the maximum Mises stress, maximum stress intensity factor, and average defect growth rate for each operating condition to construct a defect growth characteristic; and generate a defect instability growth morphology feature set based on the defect growth characteristic using a deep autoencoder and a conditional generative adversarial network;

[0289] The signal acquisition and inversion modeling module 400 is configured to collect rail strain signals through multiple FBG sensors; perform time domain, frequency domain, and time-frequency domain feature extraction on the rail strain signals to generate a first standardized feature matrix; use a convolutional neural network to construct a mapping relationship model between the first standardized feature matrix and the geometric parameter vector, and develop a convolutional neural network inversion model;

[0290] The real-time geometric description module 500 is configured to obtain the real-time rail strain signal and convert it into a second standardized feature matrix; input the second standardized feature matrix into a trained convolutional neural network inversion model to obtain a predicted geometric parameter vector and perform weighted correction; based on the weighted corrected predicted geometric parameter vector, obtain a three-dimensional predicted geometric description according to the geometric function.

[0291] In the embodiments of the present application, a system for describing the geometry of micro-defects on the tread of a high-speed train can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, the mobile electronic device can be a mobile phone, tablet computer, laptop computer, PDA, in-vehicle electronic device, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), while the non-mobile electronic device can be a server, network attached storage (NAS), personal computer (PC), etc., which are not specifically limited in the embodiments of the present application.

[0292] In the embodiments of the present application, a high-speed train tread micro-defect geometry description system can be a device having an operating system. The operating system can be an Android operating system, an iOS operating system, or other possible operating systems, which are not specifically limited in the embodiments of the present application.

[0293] The embodiment of the present application provides a high-speed train tread micro-defect geometric description system that can achieve Figure 1 In order to avoid repetition, each process of implementing a method for geometrically describing small defects on a high-speed train tread in a method embodiment will not be described here.

[0294] Optionally, an embodiment of the present application also provides an electronic device, including a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, the various processes of the above-mentioned embodiment of the method for geometrically describing minute defects in the tread of a high-speed train are implemented, and the same technical effect can be achieved. To avoid repetition, they will not be described here.

[0295] An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the various processes of the embodiment of the above-mentioned method for geometric description of tiny defects in the tread of a high-speed train are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0296] The processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes a computer-readable storage medium, such as a computer read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk.

[0297] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element. In addition, it should be noted that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved. For example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0298] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a more preferred embodiment. Based on this understanding, the technical solution of this application, or the part that contributes to the existing technology, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of this application.

[0299] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are within the protection of this application.

Claims

1. A method for geometrically describing small defects on a high-speed train tread, characterized in that: The following steps are involved: S1: Collecting morphological feature data of tiny defects on the wheel tread of a high-speed train to construct an original defect feature dataset, grouping the defects in the original defect feature dataset using aspect ratio; extracting key morphological features from each group of defects to construct a characteristic parameter matrix of the tiny defects on the wheel tread; constructing a geometric function of the tiny defects on the wheel tread using piecewise polynomial fitting and cubic spline interpolation; and constructing a standardized original defect geometry database matrix based on the characteristic parameter matrix and the geometric parameter vector in the geometric function; S2: Establish a transient finite element model of high-speed train wheel-rail rolling contact, import the geometric function into the transient finite element model, and define the wheel-rail contact relationship; use the penalty function method to define the wheel-rail contact constraint, and set the tangential friction in combination with the Coulomb friction model, and use an adaptive mesh in the defect area; Perform mechanical response simulation of the transient finite element model under different operating conditions, output the Mises stress in the defect area, and calculate the stress intensity factor using the J-integral method; S3: defining a working condition parameter matrix, wherein the working condition parameter matrix includes the geometric parameter vector; According to the stress intensity factor, the defect growth rate is approximately calculated in combination with the Paris law; For each operating condition, the maximum Mises stress, maximum stress intensity factor, and average defect growth rate are extracted to construct defect growth characteristics. Based on the defect expansion characteristics, a deep autoencoder and a conditional generative adversarial network are used to generate a defect instability expansion morphological feature set; S4: collecting rail strain signals through multiple FBG sensors; extracting time domain, frequency domain, and time-frequency domain features of the rail strain signals to generate a first standardized feature matrix; using a convolutional neural network to construct a mapping relationship model between the first standardized feature matrix and the geometric parameter vector, and developing a convolutional neural network inversion model; S5: Acquire the real-time rail strain signal and convert it into a second standardized feature matrix; input the second standardized feature matrix into the trained convolutional neural network inversion model to obtain a predicted geometric parameter vector and perform weighted correction; based on the weighted corrected predicted geometric parameter vector, obtain a three-dimensional predicted geometric description according to the geometric function.

2. The method according to claim 1, characterized in that In step S1, the morphological feature data includes the depth, area, width, length, curvature and edge gradient of the defect.

3. The method according to claim 1, characterized in that In step S1 , the geometric function is obtained by performing piecewise polynomial fitting on the defect cross-sectional profile and using cubic spline interpolation in the edge transition region.

4. The method according to claim 1, wherein In step S2, the transient finite element model is established using ABAQUS software, and the dynamic update of the defect geometry is achieved through a Fortran subroutine.

5. The method according to claim 1, characterized in that In step S3, the deep autoencoder is used to extract potential features from the defect expansion characteristics, and the conditional generative adversarial network uses the potential features and operating parameters to generate extended feature samples, and then identifies the unstable expansion pattern through a clustering algorithm to form the defect unstable expansion morphological feature set.

6. The method according to claim 1, characterized in that In step S4, the time domain features include peak strain, root mean square strain and kurtosis; the frequency domain features include peak frequency and spectral energy; and the time-frequency domain features are used to extract wavelet energy through wavelet transformation.

7. The method according to claim 1, characterized in that In step S5, the weighted correction is to perform weighted fusion on the geometric parameter vector predicted by the convolutional neural network inversion model and the cluster center of the defect instability expansion morphological feature set.

8. A high-speed train tread micro-defect geometric description system, characterized by: include: The data acquisition and preprocessing module is configured to collect morphological feature data of micro defects on the wheel tread of a high-speed train, construct an original defect feature data set, group the defects in the original defect feature data set using aspect ratio; extract key morphological features from each group of defects to construct a characteristic parameter matrix of the micro defects on the wheel tread; construct a geometric function of the micro defects on the wheel tread using piecewise polynomial fitting and cubic spline interpolation; and construct a standardized original defect geometry database matrix based on the characteristic parameter matrix and the geometric parameter vector in the geometric function; a finite element simulation module configured to establish a transient finite element model of high-speed train wheel-rail rolling contact, import the geometric function into the transient finite element model, and define the wheel-rail contact relationship; use a penalty function method to define the wheel-rail contact constraint, and set tangential friction in combination with a Coulomb friction model, and use an adaptive mesh in the defect area; Perform mechanical response simulation of the transient finite element model under different operating conditions, output the Mises stress in the defect area, and calculate the stress intensity factor using the J-integral method; a defect expansion characteristic analysis and feature set generation module, configured to define a working condition parameter matrix, wherein the working condition parameter matrix includes the geometric parameter vector; According to the stress intensity factor, the defect growth rate is approximately calculated in combination with the Paris law; For each operating condition, the maximum Mises stress, maximum stress intensity factor, and average defect growth rate are extracted to construct defect growth characteristics. Based on the defect expansion characteristics, a deep autoencoder and a conditional generative adversarial network are used to generate a defect instability expansion morphological feature set; a signal acquisition and inversion modeling module configured to acquire rail strain signals through a plurality of FBG sensors; extract features of the rail strain signals in the time domain, frequency domain, and time-frequency domain to generate a first standardized feature matrix; construct a mapping relationship model between the first standardized feature matrix and the geometric parameter vector using a convolutional neural network, and develop a convolutional neural network inversion model; The real-time geometric description module is configured to obtain a real-time rail strain signal and convert it into a second standardized feature matrix; input the second standardized feature matrix into a trained convolutional neural network inversion model to obtain a predicted geometric parameter vector and perform weighted correction; based on the weighted corrected predicted geometric parameter vector, obtain a three-dimensional predicted geometric description according to the geometric function.

9. An electronic device, characterized in that: The method comprises a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, the method implements the steps of a method for geometrically describing minute defects on a high-speed train tread as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the steps of the method for geometric description of small defects on the tread of a high-speed train are implemented as described in any one of claims 1 to 7.

Images